The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
488 
cell which has the minimum distance and updating the heap for 
every image cell addition or extraction. 
It is purposed by this storage structure to do the following big 
volume processes; reaching the image cells, testing and 
calculating the spread, marking image cells and storing the 
marked image cells. 
3.4 Conversion from raster to vector 
The algorithms of conversion from raster to vector are valid for 
the 1 bit image files which contain two types of data (0 or 1) 
(ESRI 1997). So that, in developed software, an image string is 
made that the marked pixels are full, the others are empty after 
completing the processes; selecting and marking the image cells. 
Making this image string is a masking process. This mask raster 
image is afterwards recorded as 1-bit (1 color) raster in BMP 
format. Visual C++ code (URL 1) which is available on internet 
for conversion from mask raster image to vector data is restored 
and new opportunities are added. Centers lines and border lines 
of the details converted to vectors one by one from raster data 
with a functional interface by establishing a connection with 
main program and a coordinated vector data is made by entering 
the left-bottom comer coordinates and both dimension (x,y) 
image resolutions in main program. Additionally the 
opportunity of making the vector data to the required 
smoothness by entering the tolerance is supplied in the 
functional interface. Density of the vertex points, of the vector 
data which are made by conversion from raster to vector by 
entering the smoothness tolerance, is adjustable (Eker, 2006). 
If the break point tolerance is chosen to be zero, all pixels are 
included to calculate without any smoothing. In case of the 
incidence of an increasing breakpoint tolerance, pixels with 
increasing intervals are taken into account instead of all pixels 
and the final vector is smoother. But over-increasing the 
tolerance level may cause failure on the accuracy (Eker, 2006) 
4. CONCLUSIONS 
In comparing both production systems, the recommended one 
provides approximately 5 working days in saved time. From the 
production duration point of view, this gain can not be 
neglected. 
While making feature extraction in orthophoto images and using 
the recommended production system, extraction of some 
features can be difficult, especially for some line features and 
those defined by their height values (electrical lines, towers, 
minarets etc.) which cannot be evaluated at first glance. Also, 
some features like dry streams become hard to be observed. 
Therefore, those features that are extracted on orthophoto 
images and have their height values from DEM, should 
absolutely be checked and completed by overlapping on stereo 
models. In addition to this, height errors must be corrected. As 
an example, information about the numbers of the point features 
which are evaluated just only from orthophoto image is given in 
Table 3. 
Feature 
Stereo 
Mono 
Ratio 
Building 
1142 
986 
%86 
Tree 
4796 
2571 
%53 
Bush 
336 
32 
%10 
Sheep-fold 
23 
18 
%78 
Stone 
342 
222 
%64 
Spring 
5 
1 
%20 
Culvert 
100 
60 
%60 
Fountain 
5 
2 
%40 
Water Reservoir 
6 
3 
%50 
Industrial Building 
10 
10 
%100 
Mosque 
6 
4 
%66 
Governmental 
Building 
8 
3 
%38 
School 
9 
1 
%11 
Lean-to roof 
3 
3 
%100 
Antenna 
3 
2 
%66 
Table 3: Differences which resulted from the extraction of point 
features using both stereo and mono images. 
In this study; the accuracy of the features which were digitized 
by using the suggested map production system, was investigated. 
For this purpose, 35 common Ground Control Points (GCPs) 
were selected in both maps, which were produced by using the 
existing and the suggested systems. The coordinates of the 
GCPs were measured in both maps and the RMS errors were 
calculated. At the calculation, the coordinates derived from the 
map of the existing production were accepted as reference. 
In planimetric coordinates, ± 1 meter accuracy was achieved 
while the vertical accuracy was determined as ±3 meters. The 
results of the accuracy assessment are good enough for 
1Y25.000 scaled topographic maps but it should be taken into 
consideration that these results are reliable only for this test map 
area or for similar topographies. So as to have more conclusive 
results, it would better to apply similar studies for the areas that 
have different topographic characteristics. 
Finally, it is believed that the software can be more useful for 
semi-automatic feature extraction if the deficiencies listed 
below are eliminated 
Incorrect feature extraction might occur if the appropriate 
tolerance limits are not introduced to the software. 
When the images with the very big sizes are used, software 
errors may happen because of the requirement of much memory. 
The quality of images significantly affects the performance of 
contrast and noise algorithms. 
Surface and pattern characteristics of the line features also 
effect the algorithm conclusion. 
If this software is used in the production flow then advanced 
filters like anisotropic diffusion and edge enhancement 
algorithms should be integrated. Also integrated should be 
different interpolation methods to fill the gaps that occur 
because of the obstacles. Availability of image pyramid 
methods should also be investigated to have better results. 
REFERENCES 
Eker, O., 2006. Semi-automatic extraction of line features from 
aerial photographs, Ph.D. Thesis, ITU Institute of Science, 
Istanbul. 
ESRI, 1997. ARC/INFO User’s Guide Cell-Based Modelling 
With GRID, Redlands, USA. 
Sethian, J.A., 1998. Fast Marching Methods and Level Set 
Methods for Propogating Interfaces, von Karman Institute 
Lecture Series, Computational Fluid Mechanics, Belgium. 
URL 1, 2006, Raster to Vector Transformation Program, 
http://www.xmailserver.org/davide.html